Comprehensive Visualization of Cardiac MRI Data

(1)

DISSERTATION

Comprehensive Visualization of Cardiac MRI Data

ausgeführt zum Zwecke der Erlangung des akademischen Grades eines Doktors der technischen Wissenschaften unter der Leitung von

Ao. Univ. Prof. Dipl.-Ing. Dr. techn. Eduard Gröller Institut für Computergraphik und Algorithmen

Abteilung für Computergraphik

eingereicht an der Technischen Universität Wien

Fakultät für Informatik

von

Dipl.-Ing. Maurice Alain Termeer Matrikelnummer 0827082

Lange Gasse 14 / 2 / 27 1080 Wien

Österreich

Wien, im Dezember 2008

(2)

(3)

Comprehensive Visualization of Cardiac MRI Data

Maurice Termeer

(4)

(5)

Abstract

Coronary artery disease is one of the leading causes of death in the western world. The continuous improvements in magnetic resonance imaging technology facilitate more accurate diagnoses by providing increasingly more detailed information on the viability, functioning, perfusion, and anatomy of a patient’s heart. This increasing amount of information creates the need for more efficient and more effective means of processing these data.

This thesis presents several novel techniques that facilitate a more comprehensive visualization of a patient’s heart to assist in the diagnosis of coronary artery disease using magnetic resonance imaging (MRI). The volumetric bull’s eye plot is introduced as an extension of an existing visualization technique used in clinical practice—the bull’s eye plot. This novel concept offers a more comprehensive view on the viability of a patient’s heart by providing detailed information on the transmurality of scar while not suffering from discontinuities.

Anatomical context is often lost due to abstract representations of data, or may be scarce due to the nature of the scanning protocol. Several techniques to restore the relation to anatomy are presented. The primary coronary arteries are segmented in a whole heart scan and mapped onto a volumetric bull’s eye plot, adding anatomical context to an abstract representation. Similarly, segmented late enhancement data are rendered along with a three-dimensional segmentation of the patient-specific myocardial and coronary anatomy. Additionally, coronary supply territories are computed from patient-specific data as an improvement over models based on population averages.

Information on the perfusion of the myocardium provided by MRI is typically of fairly low resolution. Using high-resolution anatomical data, an approach to visualize simulated myocardial perfusion is presented, taking full advantage of the detailed information on perfusion. Finally, a truly comprehensive visualization of a cardiac MRI exam is explored by combining whole heart, late enhancement, functional, and perfusion scans in a single visualization. The concepts introduced help to build a more comprehensive view of the patient and the additional information may prove to be beneficial for the diagnostic process.

(6)

(7)

Kurzfassung

Koronare Herzkrankheit ist eine der führenden Todesursachen in der westlichen Welt. Die kontinuierliche Verbesserung der Magnetresonanz- tomographie (MRT) erleichtert genauere Diagnosen, indem sie immer detailliertere Informationen über die Lebensfähigkeit, das Funktionieren, die Durchblutung und die Anatomie des Herzens eines Patienten liefert.

Diese zunehmende Menge an Informationen schafft die Notwendigkeit für effizientere und effektivere Mittel der Verarbeitung dieser Daten.

Diese Dissertation präsentiert mehrere neue Techniken, die eine umfassendere Visualisierung des Patienten bei der Diagnose von Erkrankungen der Herzkranzgefäße mittels MRT unterstützen. Das volumetrische Polar- diagram wird als Erweiterung des Polardiagrams, welches eine bestehende Visualusierungstechnik in der klinischen Praxis ist, eingeführt. Dieses neuartige Konzept bietet eine umfassendere Sicht auf die Lebensfähigkeit des Herzens eines Patienten, indem detaillierte Informationen über die Transmuralität der Narbe ohne Diskontinuitäten bereitgestellt werden.

Anatomische Zusammenhänge gehen in abstrakten Darstellungen von Daten häufig verloren. Darüberhinaus liefern einige Arten von Scans relativ wenig anatomischen Kontext. Mehrere Techniken zur Wieder- herstellung des anatomischen Bezugs werden vorgestellt. Die primären Koronararterien sind in einem Scan des ganzen Herzens segmentiert und werden auf ein volumetrisches Polardiagram abgebildet. Hierbei wird der abstrakten Repräsentation anatomischer Kontext hinzugefügt. Ebenso, werden segmentierte späte Anreicherungs Daten zusammen mit einer drei-dimensionalen Segmentierung des patientenspezifischen Herzmuskels und koronarer Anatomie dargestellt. Darüberhinaus werden koronare Versorgungsgebiete aus den patientenspezifischen Daten berechnet. Dies bedeutet eine Verbesserung gegenüber Modellen welche auf Bevölkerungs- durchschnitten basieren.

Informationen über die Durchblutung des Herzmuskels welche aus MRT-Aufnahmen abgeleitet werden können sind in der Regel von relativ geringer Auflösung. Unter Verwendung hochauflösender anatomischen Daten wird ein Konzept für die Visualisierung simulierter Durchblutung des Herzmuskels präsentiert. Dabei wird die detaillierte Information über die Durchblutung genutzt. Schließlich, wird eine wirklich umfassende Visualisierung einer Herz-MRT-Untersuchung erforscht. Dabei werden

(8)

Scans des ganzen Herzens, der Herzvitalität, der Herzfunktion und der Durchblutung in einer einzigen Visualisierung kombiniert. Die einge- führten Konzepte fördern den Aufbau eines umfassenderen Überblicks über den Patienten. Die dabei zusätzlich gewonnene Information kann für den Diagnoseprozess von Nutzem sein.

(9)

Preface

T

his work would not have been possible without the help and support of many people. I would like to thank Meister Eduard Gröller, Javier Oliván Bescós, and Marcel Breeuwer for sharing their knowledge, expertise, and inspiration and for the many fruitful discussions we have had. Furthermore I would like to thank Anna Vilanova and Frans Gerritsen for their scientific contributions and additional support, Eike Nagel for his contributions as a medical expert, and the people of the Philips Healthcare, Healthcare Informatics, Clinical Science & Advanced Development group.

Last but not least I would like to express my thanks for providing a motivating and fun working environment to the people of the Visualization Group of the Vienna University of Technology, including Stefan Bruckner, Raphael Fuchs, Martin Haidacher, Peter Kohlmann, Muhammad Muddassir Malik, Matej Mlejnek, Daniel Patel, Peter Rautek, Ivan Viola, and Erald Vuçini.

This work was performed in the scope of the COMRADE project¹ funded by Philips Healthcare, Best, The Netherlands. Several datasets used during this project were provided by Hyogo BHC at Himeji, Japan, the Tokyo Metro Police Hospital, Japan, the Medical Satellite Yaesu Clinic, Japan, and the Kumamoto Cyou Hospital, Japan.

Vienna, Austria, December 2008 Maurice Termeer

1http://www.cg.tuwien.ac.at/research/vis/comrade

xiii

(14)

(15)

Experience is what you get if you didn’t get what you wanted.

Dan Stanford

Introduction 1

Heart disease is one of the leading causes of death in the western world [4]. Most patients, especially at later age, suffer from coronary artery disease. The continuous improvement in tomographic imaging technology is enabling the early detection and accurate diagnosis of this disease. With a continuously increasing amount of data produced by modern scanning technology, there is a need for more efficient and effective methods for processing these data. This chapter gives a brief introduction into the anatomy of the human heart, coronary artery disease, how magnetic resonance imaging can be used to diagnose this disease, and currently common approaches to process and visualize the resulting tomographic cardiac data.

1.1 The Human Heart

T

he human heart is one of the first organs to form in a human embryo.

It maintains the circulation of blood throughout the body from three weeks after conception and continues to do so until the very end of a human life. This muscular organ is located slightly left of the center of the thorax, beneath the breastbone and is surrounded by the lungs.

The majority of the heart is formed by a body of muscle tissue called the myocardium. The myocardium is surrounded by the pericardium on the outside and the endocardium on the inside of the heart. Figure 1.1 depicts an annotated anterior view of a human heart, while Figure 1.2 show the inside of the heart.

A human heart can be divided into a left and a right part. Each half consists of an atrium and a ventricle, giving the heart four chambers in total. The left and right halves are separated by the septum. The right

1

(16)

Figure 1.1: Anterior view of the heart depicting the atria, ventricles, connecting vessels, and the coronary arteries. Image courtesy of Gray [73].

heart is responsible for circulating blood through the lungs, while the left heart supplies the remainder of the body with oxygen-rich blood. As the left ventricle has to pump blood through a much larger part of the body than the right ventricle, the left ventricle has a thicker, stronger muscular wall. The tip of the left ventricle is called the apex, while the area near the top is called the base.

The repeating process of pumping blood through the heart by a series of muscle contractions is called the cardiac cycle. This cycle can be divided into two phases. The first phase is when the ventricles are in a relaxed state and are being filled with blood. This phase if called diastole. During the second phase the ventricles contract to pump out blood. This phase is called systole.

Each cardiac cycle, oxygen-depleted blood flowing through the inferior and superior vena cava enters the heart in the right atrium. It is then pumped through the tricuspid valve into the right ventricle. From there it flows through the pulmonary valve into the pulmonary artery, which leads the blood through the lungs. Inside the lungs gas exchange between carbon dioxide and oxygen occurs. The now oxygen-rich blood then flows through the pulmonary vein into the left atrium. Passing through the mitral valve, it enters the left ventricle, which finally pumps it through the aortic valve into the aorta.

Near the beginning of the aorta, two smaller vessels branch off. These are called the left and right coronary artery. These coronary arteries branch into a network of smaller arteries that supply the heart with oxygenated

(17)

Coronary Artery Disease 3

Figure 1.2: Anterior view of the heart exposing the inside of the ventricles, the valves, and the septum. Image courtesy of Gray [73].

blood. In Figure 1.1 the right coronary artery and left anterior descending are well visible. These two arteries are often abbreviated RCA and LAD, respectively. On the posterior side of the heart there also is the left circumflex, another branch of the left coronary artery often abbreviated LCX. These three arteries are the three main coronary arteries. They each branch into a network of smaller vessels. The exact shape and location of the coronary artery network differs greatly among individuals.

1.2 Coronary Artery Disease

Coronary artery disease is a collection of diseases related to a deficiency in one or more of the coronary arteries. It is sometimes also referred to as coronary heart disease and is commonly abbreviated CAD or CHD.

Coronary artery disease is caused by the accumulation of atheromatous plaques on the walls of the coronary arteries. These plaques may form due to a number of reasons, including smoking, diabetes, high cholesterol, old age, and obesity. Over time a plaque may calcify and remain within the vessel wall. These plaques are relatively harmless since they do not cause a significant narrowing of the vessel. More harmful are lipid core plaques, which do not calcify. These plaques gradually increase in size and expand into the lumen of the vessel. Due to inflammation they can eventually rupture, which may obstruct the blood flow in the vessel. The acute rupture of a plaque is a common cause of sudden death, usually due to an infarction of the myocardium. Such an infarction is however not necessarily fatal. Plaques extending into the lumen of the vessel can also cause problems without rupturing. They may cause a partial or complete

(18)

obstruction of the vessel known as a stenosis. It can take several years before the effects of such an accumulation of plaque becomes noticeable.

Both an acute and chronically developed obstruction of a coronary artery can trigger a myocardial infarction, an event also known as heart attack. During this event myocardial tissue becomes infarcted due to a lack of supply of oxygenated blood. Without a quick response, the tissue in the affected area will receive irreversible damage. After an infarction the dead tissue will turn into scar. If the infarcted area does not extend through the entire wall of the ventricle, the healthy part of the wall can thicken while the scar thins over time. This degree of how far scar extends into the wall is called the transmurality of the scar.

Within the area of coronary artery disease, the following three types of diseases, each caused by a partial or complete obstruction of the blood flow in one or more of the coronary arteries, are commonly distinguished.

Ischemia or ischemic heart disease is a condition where part of the myocardium does no longer receive sufficient oxygenated blood to contract, but is not yet infarcted, i.e. the muscle tissue is still alive.

Tissue with reduced perfusion is said to be hypoperfused, which can also occur without a decrease in contractility. When contractility is decreased, this type of tissue is commonly called hibernating myocardium. Reduced wall contraction lowers the cardiac output, an indicator for how much blood the heart can pump. Ischemia can be caused by both partial and complete obstructions of a vessel. Restor- ing the proper supply of oxygenated blood through revascularization will restore the proper functioning of the myocardium.

Chronic infarction refers to a condition where the patient has already experienced one or more myocardial infarctions. An infarction occurs when part of the myocardium receives too little oxygenated blood to remain viable and is more commonly caused by complete than partial obstructions of a vessel. The muscle tissue dies and proper functioning of this tissue can no longer be restored. Over time the dead tissue will turn into scar. Myocardial tissue surrounding an area of scar can be ischemic. This tissue may show reduced wall contraction, but is not yet scar and can thus still benefit from restoring the blood flow by revascularization.

Acute infarction refers to an acute rupture of a plaque, causing a sudden block of oxygenated blood and preventing part of the myocardium to contract. Without a supply of oxygenated blood, infarction will occur after about 20 minutes. If the blood flow can be restored quickly enough, the affected area, called the area at risk, may still recover.

The time frame in which revascularization is considered to be useful is around six hours.

(19)

Magnetic Resonance Imaging 5 The care cycle of coronary artery disease starts with a proper diagnosis.

If it is concluded that the patient would benefit from intervention, several techniques exist for restoring the blood flow, a process known as revascularization. Bypass surgery consists of grafting vessels from another part of the body to bypass the obstructed part of a vessel. Coronary angioplasty involves placing a small balloon inside the obstructed vessel and then inflating it to widen the vessel. A stent can be placed to maintain the vessel shape after the balloon is removed. Finally, a blood cloth causing obstruction can be diluted through pharmacological means by a technique called thrombolysis.

1.3 Magnetic Resonance Imaging

Magnetic resonance imaging (MRI) is a non-invasive [23] tomographic imaging technique based on spin properties of hydrogen nuclei. While a brief introduction into the basics of MRI follows, the reader is directed to other sources for a more thorough description of the technology behind MRI [32, 34].

An MRI scanner generates a strong, static magnetic field. The common field strength in current clinical practice is1.5 Tesla, although a transition towards3.0 Tesla is in progress [27, 58]. The organic compounds found in a human body contain relatively many hydrogen nuclei. These subatomic particle possess the quantum mechanical property of spin. When a body is placed inside the scanner, the spins of these hydrogen nuclei align themselves to the magnetic field in either a low or a high energy state.

Under normal conditions, there are slightly more nuclei in the low energy state than in the high energy state. Nuclei in the low energy state can absorb a photon that has an energy that matches the difference in energy between the low and high energy states. Thus a radio frequency pulse can be used to excite the nuclei in the low energy state and throw them in the high energy state. This phenomenon is called resonance and explains the name magnetic resonance imaging. The exact frequency of the required radio frequency pulse is called the Larmor frequency and depends on many factors, including the strength of the magnetic field.

When the radio frequency pulse is stopped, the ratio between nuclei in the low and high energy states will restore itself. During this process nuclei that jump from the high energy state back to the low energy state emit the previously absorbed energy in the form of a radio frequency wave. It is this energy that can be detected using receiver coils, giving the MRI signal. Since different tissues have a different proton density, the MRI signal they cause also differs. Besides the proton density, two other tissue-specific characteristics can be read from the MRI signal. These two properties are related to the net magnetization of the nuclei. The

(20)

first is the rate at which the equilibrium of energy states is restored, the spin lattice relaxation time or T1. This is the time for the longitudinal magnetization, the component aligned with the magnetic field, to increase by a factor ofe. The second is the rate at which the spins of the nuclei dephase, the spin-spin relaxation time or T2. This is the time for the transverse magnetization, the component orthogonal to the magnetic field, to decrease by a factor ofe. Inhomogeneities in the magnetic field of the scanner affect the measurement ofT2. The measured decay factor, which is always less thanT₂, is referred to asT^∗₂.

Three magnetic gradients inside the scanner can locally change the magnetic field strength. Since the Larmor frequency depends on magnetic field strength, these gradients can be used to control which nuclei react to a radio frequency pulse. A slice selection gradient defines an imaging plane in which the nuclei react. A phase-encoding and frequency-encoding gradient are then used to further control the way the nuclei within the imaging plane react to the radio frequency pulse. The type of pulse used and the rate at which it is repeated affect the contribution of proton density, T₁ and T₂ to the final image. A phase-frequency image is constructed and using an inverse Fourier transformation, a two-dimensional image visualizing the anatomy can finally be obtained.

The main advantages of MRI compared to other imaging modalities are that it offers excellent soft tissue contrast and a high spatial and temporal resolution on an imaging plane of any orientation with a good signal- to-noise ratio. Since MRI uses harmless magnetic fields, it is also well suited for screening purposes. This is less true for some other tomography techniques. Computed tomography (CT) is based on X-ray technology and thus uses harmful ionizing radiation. Nuclear medicine methods including single photon emission computed tomography (SPECT) and positron emission tomography (PET) use harmful radioactive compounds.

Disadvantages of MRI include the relatively high cost of the scanner.

Furthermore since an MRI tube is rather narrow, some patients may experience claustrophobia. Also the presence of ferro-magnetic implants in a patient can be a reason not to undergo an MRI exam [23], although the acceptance of patients with pacemakers is being reconsidered [52].

1.3.1 Cardiac MRI

MRI is a powerful diagnostic imaging technique for the diagnosis of coronary artery disease [47, 19, 66, 69]. MRI in the context of diagnosing heart diseases is commonly called cardiac MRI. Given a patient with suspected coronary artery disease, making an accurate diagnosis and adequate treatment planning requires information on several aspects of the heart. MRI can provide information on the contractile functionality of the myocardium, the viability of myocardial tissue, changes in the perfusion

(21)

Magnetic Resonance Imaging 7 of the myocardium and changes in the myocardial and coronary arteries anatomy. These are all important factors for the diagnosis of coronary artery disease. This diversity is one of the main strengths of cardiac MRI.

The sections below give a brief introduction to the individual cardiac MRI scanning protocols that are used to assess these properties.

Although MRI is an attractive technique for the diagnosis of coronary artery disease, in current clinical practice echocardiography, X-ray angiography, and nuclear medicine are still more prevalent than both MRI and CT. Reasons include MRI being rather expensive and relatively young.

Good correlations between MRI perfusion imaging, X-ray angiography, and SPECT have been reported [81, 55] and the popularity of MRI in cardiac medicine is expected to increase.

1.3.2 Whole Heart Imaging

Whole heart imaging [88] provides a high resolution scan of the cardiac anatomy. A typical resolution for a whole heart scan is 160 slices of 512 by 512 voxels each. While it reveals little about the actual functioning of the heart, it provides detailed anatomical information on the shape and size of the ventricles and the location of the coronary arteries. Current whole heart MRI technology is however not yet capable of imaging the coronary arteries at a quality level that can compete with X-ray coronary angiography or coronary CT. Magnetic resonance angiography (MRA) is a related scanning protocol specifically for imaging the coronary arteries [9, 71, 90]. Since the coronary arteries can be hard to detect in a regular whole heart scan, a contrast agent is occasionally used to emphasise the coronary arteries.

While the current clinical standard remains X-ray angiography, the less invasive nature and good correlations of MRA to X-ray angiography make it an attractive alternative angiography technique [1, 17, 41]. However, due to its higher spatial resolution, coronary CT angiography is also becoming more prevalent [33].

An example of two reformats from a whole heart scan are shown in Figure 1.3. This figure depicts two common orientations of the imaging plane in cardiac tomography. In the short-axis view (see Figure 1.3a) the imaging plane is perpendicular to the long axis, one of the two standard cardiac axes that runs from the base of the left ventricle to the apex. The 4-chamber long-axis view shown in Figure 1.3b is aligned to both the long and short axes. The latter axis runs approximately from the mitral valve to the tricuspid valve.

Due to its relatively high resolution a whole heart scan is well suitable for segmentation purposes [20]. A high quality segmentation is useful for visualization purposes, as will be demonstrated in this thesis. Due to continuous improvements of MRI technology [82, 38], the additional

(22)

(a) (b)

Figure 1.3: Two reformatted images from a whole heart scan showing (a) a short-axis view and (b) a 4-chamber long-axis view.

scanning time required for a whole heart scan is decreasing to a single breathhold.

1.3.3 Functional Imaging

Directly assessing the functionality of the heart provides information on whether any pathology already has had an effect. Capturing this type of information is possible with functional imaging, also called cine MRI.

A cine scan provides dynamic image data of a beating heart. It has proven to be an effective method for detecting abnormalities in cardiac functionality [86, 54, 56] and has been shown to have increased sensitivity and specificity compared to echocardiography [68]. Common exams consist of the acquisition of a stack of short-axis slices, a 2-chamber long-axis view, and a 4-chamber long axis view. Figure 1.4 shows three short-axis slices from a cine-scan at the same slice scanning location but at different phases of the cardiac cycle. In order to deal with cardiac motion, an electrocardiogram (ECG) is used to determine the phase of each slice during scanning. Once all slices are acquired, this information can be used to place the slices in the correct order. Current MRI technology allows for the acquisition of a cine-sequence of a full heart beat in a single breathhold, but due to time restrictions such scans are typically limited to three to five slices [47]. Free-breathing cine allows for the acquisition of more slices through the use of respiratory gated scanning techniques, where in addition to an ECG also the lung-liver interface is monitored

(23)

Magnetic Resonance Imaging 9

(a) (b) (c)

Figure 1.4: Three short-axis slices from a functional scan of different phases of the cardiac cycle ranging from (a) end-diastole to (c) end-systole.

to ensure all slices are scanned at approximately the same phase of the respiratory cycle.

A functional scan can be quantitatively analyzed to provide absolute measurements on the functioning of the heart [67]. Quantitative analysis requires a segmentation of the myocardium. Automatic segmentation techniques can greatly reduce the required amount of work of making this segmentation [29]. Several measures are commonly used to assess the relative health of a patient’s heart. Stroke volume is the difference in volume of the left ventricular blood pool between end-diastole and end- systole. Ejection fraction is the stroke volume divided by the end-diastolic volume. Cardiac output is the stroke volume multiplied by the heart rate.

Wall thickening is the difference in wall thickness at end-diastole and end-systole divided by the end-diastolic wall thickness [11]. The results of a quantitative analysis can be presented graphically by first dividing the myocardium in a number of segments on each slice. These segments can then be visually arranged into a single diagram called a bull’s eye plot, showing a quantitative measure, for example wall thickening, for the entire left ventricle in a single image. The bull’s eye plot is further discussed in Section 1.4.

While an MRI exam virtually always contains a functional scan at rest, functional scans at different stress levels can also optionally be acquired.

A common stress inducing agent is dobutamine [86, 44]. The type of exam to measure function under stress using this contrast agent is often called dobutamine stress MRI. At a higher stress level the myocardium requires more oxygen. Any decreased functionality due to an obstructed supply of oxygenated blood may therefore be more pronounced at higher stress levels.

(24)

Figure 1.5: A slice from a late enhancement scan. The arrow indicates an area showing hyperenhancement near the endocardium.

1.3.4 Late Enhancement Imaging

Myocardial tissue that has died due to a prolonged lack of supply of oxygenated blood can be detected using late enhancement imaging, also referred to as delayed enhancement or delayed contrast enhancement imaging [89, 42, 16, 12, 8]. A contrast agent, most often gadolinium- based, that shows up bright on an MRI image is injected into the patient.

Approximately 15 to 20 minutes after injection a scan is made. A scan typically consists of 15 to 20 short-axis slices, a 2-chamber long axis slices, and optionally a 4-chamber long-axis slice. All slices are taken at approximately the end diastolic phase of the cardiac cycle. An example of a short-axis late enhancement slice is shown in Figure 1.5. The injected contrast agent accumulates in infarcted tissue, while washing out of the well-perfused, viable parts of the myocardium. In the resulting scan hyperenhancement, an increase in signal strength due to contrast agent accumulation, is therefore observed in infarcted tissue. Some other tissues, such as the fat surrounding the heart, also show up bright in a late enhancement scan.

When assessing viability, the location and size of the infarcted region are relevant attributes to determine. The degree to which an infarcted region covers the myocardial wall, the transmural extent of infarction, is also important [15]. Scar can occur for example only near the endocardium, only near the epicardium, or in the middle of the myocardial wall. The exact location provides information on the nature of the infarct. If there is sufficient healthy tissue remaining, the infarcted tissue may thin over time and the wall can recover. Obtaining detailed information on the

(25)

Magnetic Resonance Imaging 11 transmurality of scar as well as the absolute amount of healthy tissue remaining is thus important for making an accurate diagnosis.

Also a late enhancement scan can be quantitatively analyzed. A common approach consists of first segmenting the myocardium. Since the scan consists of relatively few slices and only a single time step, a manual segmentation approach is feasible. Similar to the quantitative analysis of functional data, also this segmentation can be used to divide the myocardium in a number of segments. For each segment the percentage of scar and the transmural extent of infarction can then be visually presented in a bull’s eye plot.

1.3.5 First-Pass Perfusion Imaging

Decreased contractility and scar are both the results of a reduced oxygen supply to the myocardium. The perfusion of the myocardium itself can also be visualized using first-pass perfusion imaging [55, 72, 2, 13, 35].

Similar to late enhancement imaging, this protocol relies on the use of a contrast agent. Instead of letting the contrast agent accumulate, the first pass of the contrast agent through the heart is captured. Since this imposes severe timing constraints—the scan has to be completed after the contrast agent has passed through the heart—the number of slices is typically limited to three to five at a temporal resolution of 10 to 30 time steps. The scanner, the patients heart rate, and the duration a patient is capable of holding his breath all influence these limitations. A perfusion scan is always scanned under artificially induced stress. In contrast to functional imaging, a vasodilator, e.g. adenosine, is used to widen the vessels, simulating the effects of stress while keeping the heart rate low.

Optionally a rest perfusion scan is made after the stress perfusion scan.

Figure 1.6 shows a few slices from a first-pass perfusion scan. A proper perfusion scan shows the contrast agent first enter the right ventricular blood pool, then enter the left ventricular blood pool and finally diffuse into the myocardium. In the latter phase a perfusion defect manifests itself as an area that lights up less bright—or not at all—than the rest of the myocardium. This is indicated by the white arrows in Figure 1.6d.

Due to the breathing motion from the patient during scanning, a perfusion scan often suffers from motion artifacts. Therefore the quantitative analysis of a perfusion scan often requires a registration between the various time steps of each slice. Note that registration can only compensate for motion parallel to the imaging plane. After the myocardium is segmented in the registered scan, a time-intensity curve can be constructed for each voxel of the scanned myocardium. Many parameters can be derived from this curve, including the average speed with which the signal strength increases, the time it takes to reach maximum signal strength and the maximum signal strength itself. These values are known as mean or maxi-

(26)

(a) (b) (c) (d)

Figure 1.6: Four short-axis slices from a first-pass perfusion scan showing (a) no enhancement, (b) the contrast agent entering the left ventricular blood pool, and (c) & (d) diffusing into the myocardium. The darker area in the myocardium indicates a perfusion defect.

mum upslope, time-to-peak, and peak enhancement, respectively. When both a stress and rest scan are available, the myocardial perfusion reserve index (MPRI) can be computed, which is the ratio between the maximum upslope of the stress and rest scan. This measure has been shown to be an accurate indicator for coronary artery disease [55].

1.4 Visualization of Cardiac MRI Data

Visualization techniques can help reduce the workload by presenting data in a more accessible manner. Much work has been done on visualization techniques for cardiac MRI data. In this section a few of the most commonly applied techniques are mentioned. The remaining chapters contain a more complete discussion of relevant related work.

In current clinical practice, the primary methods of analyzing data from an MRI exam are based on visual inspection of the reconstructed slices. This requires much expertise from the clinician, for example in discriminating scanning artifacts from pathologies. It also means that combining the information provided by the scans of the various protocols has to be done mentally, by examining scans successively or side-by-side.

Segmentation is being applied increasingly, as it not only facilitates taking quantitative measurements but also creates possibilities for more advanced visualization techniques. Segmenting the left ventricular myocardium in a functional scan enables the computation of local wall thickness and subsequently related measurements including wall thickening, stroke volume, and ejection fraction. Segmenting the left ventricular myocardium in a late enhancement scan allows to depict the viability of the entire left ventricle in a two-dimensional schematic visualization called a bull’s eye plot. Figure 1.7 shows two bull’s eye plots of late enhancement data. Each concentric ring depicts data from one short-axis slice. The

(27)

Visualization of Cardiac MRI Data 13

0%

17 33 50 67 83 100%

(a)

4 3 2 1 0

(b)

Figure 1.7: Two bull’s eye plots of a late enhancement scan representing (a) percentage of scar and (b) transmurality.

quantitative analysis results of both functional and perfusion scans are also commonly depicted in bull’s eye plots. A more thorough description of the bull’s eye plot is given in Chapter 2.

In a whole heart scan the individual parts of the heart can be segmented fully automatically [20]. This provides a patient-specific model of the heart, suitable as anatomical context in visualizations. The centerlines of the coronary arteries can also be segmented in a whole heart scan [50].

Current MRI scans do not yet offer sufficient level of detail to provide accurate vessel diameter measurements. Since the coronary arteries have a complex shape, examining them in two-dimensional slices is a tedious task that requires a lot of training. Once segmented, the coronary arteries can be visualized while preserving their complex shape by reformatting the whole heart scan on a three-dimensional surface that intersects the coronary arteries [21, 83]. Abstracting from their complex shape by virtually stretching the arteries and visualizing the reformatted whole heart scan has the advantage that the image data covering the entire artery can be observed in two dimensions [39].

The abundance of information provided by the available scanning protocols is one of the main advantages of cardiac MRI over other tomographic imaging technologies. Combining the information contained within the data of the multiple scans provides a more detailed under- standing of the patient and allows for a more accurate diagnosis. It has been shown that combining perfusion and viability information [12, 30], optionally with coronary arteries [31], or combining functional and viability information [59, 26], provides an accurate discrimination between viable, ischemic, and infarcted tissue and also facilitates the detection of hibernating myocardium.

(28)

1.5 Scope of this Thesis

The large amount of mental processing involved in current diagnostic work flows seems to be partially caused by not explicitly visualizing relations between different types of data. This thesis explores several visualization approaches that present various types of data in a rich anatomical context. Several data types can be combined to increase spatial correlation between them. Data can also be represented differently, to easier extract desired information. Together these visualization approaches aim towards a comprehensive visualization of cardiac MRI data, combining the anatomy of the myocardium and the coronary arteries with various measurements about viability, function, and perfusion of the myocardium.

Chapter 2 introduces the volumetric bull’s eye plot, an extension of the bull’s eye plot that is continuous, preserves the volumetric nature of the myocardium, and adds anatomical context to the otherwise rather abstract representation of the left ventricle. This technique also provides a much more detailed visualization of transmurality of scar. This work was presented at IEEE Visualization 2007 [79] and the SCMR 11th annual scientific sessions [77]. Patent applications on this concept were filed under application number PH-008909-EP-1 (ID 680295) and PCT/IB2008/053464.

Chapter 3 presents the visualization of viability in a three-dimensional anatomical context, providing a more detailed relation between scar and coronary arteries than is common in previous methods. This work was presented at IEEE Visualization 2007 [79].

Chapter 4 investigates computing patient-specific coronary territories, i.e., computing which parts of the myocardium are supplied by which coronary arteries based on anatomical data from the patient. This work was filed as a technical report [75] and accepted for a poster publication at the SCMR 12th annual scientific sessions [78]. A patent application on this concept was filed under application number PH-010411-EP-1 (ID 676667).

Chapter 5 shows how ideal perfusion data could be visualized. The input for these methods is given by a simulation of myocardial perfusion based on patient-specific anatomy. Using simulation data provides several interesting new possibilities for visualization and also provides another detailed visualization of the patient’s anatomy. This work was presented at IEEE VisWeek 2008 [80].

Chapter 6 discusses an approach to provide a comprehensive view of a patient by integrating anatomical, functional, viability, and perfusion data into a single image.

Finally, Chapter 7 presents a summary and the conclusion of the work in this thesis.

(29)

Any sufficiently advanced technology is indistinguishable from magic.

Arthur C. Clarke

The Volumetric Bull’s Eye Plot 2

The bull’s eye plot is an accepted medical visualization technique commonly used during the diagnosis of patients with suspected coronary artery disease. It provides a visual representation of a quantitative measure for the entire left ventricle in a two-dimensional diagram. This chapter presents an extension of this technique in the context of assessing viability using a late enhancement cardiac MRI scan. The extended concept is based on an unfolding the left ventricular myocardium along the long axis. The improved diagram does not suffer from discontinuities, preserves the volumetric nature of the left ventricular wall, and adds anatomical context. The preservation of wall thickness facilitates a direct and detailed visualization of the transmurality of scar. These improvements provide a more comprehensive view of the viability of the left ventricular myocardium of a patient in a way that can more easily be related to the patient-specific anatomy.

2.1 Introduction

T

he detection and quantification of infarcted myocardial tissue, also called scar, is of importance for the assessment of the infarct location and severity. The severity of scar is partially determined by the transmurality, a measure for how far the scar extends into the left ventricular wall. Quantification of scar is also important for the selection and planning of therapy such as revascularization via the placement of a stent or a bypass.

Late enhancement imaging is an MRI protocol suitable for the assessment of myocardial viability [89, 8]. It uses a contrast agent to visualize

15

(30)

scar. A late enhancement scan is made 15 to 20 minutes after injection of a contrast agent. Because scar can contain more contrast agent than healthy tissue and experiences a delayed wash in of the contrast agent, it shows up brighter in the scan than healthy tissue. For each MRI scanning protocol, the scanning time is minimized such that the desired information for that protocol is still obtained. Therefore the atria are for example not covered by a late enhancement scan, as they are not relevant for the assessment of viability. Late enhancement scans commonly consist of much fewer slices than whole heart scans, although the in-plane resolutions of both protocols are approximately equal. Due to the scanning protocol used, the coronary arteries are often not well visible. Late enhancement scans thus exhibit less anatomical detail than an whole heart scan.

In current clinical practice the results of MRI-based scar quantification methods are examined using solely two-dimensional visualization techniques. Apart from examining the individual slices, bull’s eye plots are often used to obtain a schematic representation of the scar in a patient. A bull’s eye plot maps the left ventricular wall to a set of concentric rings, the middle ring corresponding to the apex and the outer ring corresponding to the base. This two-dimensional visualization technique gives an overview of the entire late enhancement scan and eliminates shape variations of the left ventricle between patients. A common approach for assessing the left ventricular viability using bull’s eye plots is to show several dia- grams simultaneously, for example one showing the relative amount of scar and another showing the transmurality of scar. The bull’s eye plot has two important drawbacks. First it is difficult to make a relation to the three-dimensional anatomy. Even when using the late enhancement slices, all relations have to be constructed mentally. Second it is impossible to relate scar to specific coronary arteries—which may be the cause of the scar—in a patient-specific way, because those are not well visible in a late enhancement scan.

This chapter presents the volumetric bull’s eye plot, a novel visualization technique for the diagnosis of patients with coronary artery disease, that overcomes the aforementioned limitations. This technique extents the bull’s eye plot by unfolding the myocardium while preserving the volumetric nature of the left ventricular wall. This approach preserves continuity, is better scalable in the number of slices of the late enhancement scan, and allows for a more detailed visualization of transmurality. The latter advantage may allow for a more detailed assessment of the severity of scar. Finally, the diagram is annotated with anatomical context to allow for a better relation to anatomy and other tomographic data of the patient.

This chapter is structured as follows. Section 2.2 discusses related work. Section 2.3 discusses the improved visualization techniques in detail.

Finally, Section 2.4 concludes this chapter.

(31)

Related Work 17

8 7

9 10 11 mid- 12 cavity

13 14

15 apical16 17

basal

apical cavitymid-

2 1

3 4 5

basal 6

(a)

1 2

3

4

5 7 6

8

9

10

11 13 12

14 15

16 17

1 2

3

4

5 7 6

8

9

10

11 13 12

14 15

16 17

LAD RCA LCX

(b)

Figure 2.1: The 17-segment model as proposed by the American Heart Association [14]: (a) The division of the left ventricular myocardium into 17 segments; (b) the 17 segments displayed in a bull’s eye plot.

2.2 Related Work

The American Heart Association has developed a standardized model for dividing the myocardium into segments for use among various tomography techniques in clinical practice [14]. The model aims at greatly increasing the comparability and reproducibility of diagnoses. The model is illustrated in Figure 2.1. It divides the left ventricle into a basal, a mid-cavity, and an apical part along the long axis. These three parts are further divided into segments, as is shown in Figure 2.1a. A separate segment is allocated to the apex. The model also suggested a mapping between these segments and the coronary arteries, shown in Figure 2.1b. This mapping is based on population averages. Later studies have investigated this mapping [65, 64]

and have found that it should be used with care, as there are large variations in the anatomy of the coronary arteries among individuals.

2.2.1 The Bull’s Eye Plot

Besides the segmentation model, the American Heart Association also presented a schematic approach to represent the left ventricle in a single diagram. This diagram is called a bull’s eye plot [36]. It is a mapping of the myocardium to a set of concentric rings. The bull’s eye plot in the model proposed by the American Heart Association consists of three rings, but the concept can be extended to any number of rings. Each ring

(32)

represents the segmented left ventricular myocardium on one slice of a short-axis scan. The apex, represented by the middle of the bull’s eye plot, is often represented by a separately scanned long-axis slice. For each slice of the short-axis late enhancement scan, the left ventricular myocardium is reformatted to a ring. The inner ring represents the region near the apex, the bottom of the left ventricle, while the outer ring represents the area near the base, the top of the left ventricle. Each ring is divided into a number of segments, which can subsequently be colored according to a measured property. While the bull’s eye plot as a visualization concept can be applied to various types of measurements, this chapter focuses on viability information provided by a late enhancement scan. An example of a bull’s eye plot of late enhancement data is shown in Figure 2.3a. The bull’s eye plot is popular in medical practice as it is intuitive and gives a standardized, global overview of the property being measured.

The left ventricular myocardium, apart from the apex, has approximately the shape of a cylinder. It is however not perfectly circular and the wall thickness varies throughout the left ventricle. These properties, as well as the general shape, size, and orientation, differ for each patient. The bull’s eye plot was developed to allow for a structured and reproducible analysis that is easily comparable between different patients, or different scans of the same patients, by eliminating all shape variations.

Although the bull’s eye plot eliminates all the inter- and intra-patient shape variations of the left ventricle, it has three main drawbacks. First the quantization of the data into segments gives a less accurate view than a continuous one. Implementations exist that provide more detail by using a high number of segments or interpolation. However, the bull’s eye plot is discontinuous at the borders between the rings of the different slices. The second, more severe drawback is that it provides no information on the distribution of scar tissue throughout the ventricle wall—the transmurality of scar. This problem is sometimes solved by displaying two bull’s eye plots simultaneously: one that visualizes the amount of scar and one that visualizes the transmurality of scar. The third drawback is that it provides very little anatomical information. The bull’s eye plot is a two-dimensional, abstract representation of the late enhancement data, but relating it to anatomical information is necessary. Even with an additional view that provides anatomical information, the relation between these two views would have to be done mentally.

2.2.2 Combinations with Other Visualization Techniques

Noble et al. [59] combine late enhancement data with information on myocardial contraction to identify regions of the myocardium that suffer from ischemia but are not yet scarred. Such regions can still be revived and identification thereof is thus useful. They propose both the use of a bull’s

(33)

Related Work 19 eye plot and a projection onto the left ventricular wall as visualization of the resulting parameters. The latter aims to visualize the measurements in an anatomical context, since the bull’s eye plot is an abstract representation without anatomical information. In their work the results were limited to providing shape information of the left ventricular myocardium, since only a segmentation of the left ventricle was available.

Besides the bull’s eye plot, more visualization methods for the analysis of cardiac tomography data have been proposed, although most of these techniques have not yet been accepted in medical practice. Kuehnel et al. [43] designed software assistants for the analysis of cardiac CT and MRI data. Using segmentations of the coronary arteries and the left ventricle, they provide orthogonal cross-sections on the coronary arteries to inspect vessel diameters, bull’s eye plots and color-coded slices of perfusion data, and an integrated visualization showing a three-dimensional left ventricle with perfusion data and the coronary arteries. In the same context Oeltzeet al.[62] presented various visualization techniques for the analysis of perfusion data. There an integration of visualizations of perfusion measurements with anatomical information is proposed. This allows perfusion defects to be easily related to coronary artery branches feeding the affected area. Both these approaches propose the use of schematic visualization techniques such as the bull’s eye plot in combination with three-dimensional visualization techniques to visualize the data in an anatomical context. A limitation of both approaches is that the schematic visualizations still contain fairly little anatomical detail themselves. The separate visualizations therefore still need to be combined mentally to put the schematic data into anatomical context.

2.2.3 Segmentation of the Myocardium

The construction of a bull’s eye plot—or a volumetric bull’s eye plot—

requires the myocardium to be segmented in the late enhancement images.

In this work manual contouring was used. Endocardial and epicardial contours were drawn by the user on each short-axis late enhancement slice where these structures are visible. An example of a late enhancement slice with manually drawn contours is given in Figure 2.2. For the volumetric bull’s eye plot these contours are then interpolated using uniformly sampled Catmull-Rom splines to form a three-dimensional quadrilateral mesh. This mesh, which typically does not include the apex, forms a segmentation of the relevant parts of the left ventricular myocardium.

Besides a segmentation, a method is needed to discriminate scar from healthy tissue. This is commonly accomplished with the definition of image intensity ranges for both scar and viable tissue. In this work user- defined ranges are used. Inside the manually drawn contours, the user specifies two regions; one that clearly represents healthy tissue and one

(34)

Figure 2.2: A late enhancement slice with manually drawn contours on the endocardium (green) and epicardium (yellow), healthy region (blue), and scar region (red).

that clearly represents scar. These two regions are shown in blue and red respectively in Figure 2.2. For both regions,[µ−2σ,µ+2σ] is used as the intensity range that represents either healthy tissue or scar. In this formula µ denotes the mean intensity and σ denotes the standard deviation. In case these two intervals do not overlap, a linear transition is defined on the interval[µ_healthy+2σ_healthy,µ_scar−2σ_scar]. For some cases with little scar or good image contrast between scar and viable tissue a fuzzy thresholding approach was used instead. Here the user interactively specifies values µ_mid and σ_mid. Then all intensity values below µ_mid−σ_mid are considered viable tissue, all intensity values above µ_mid+σ_mid are considered scar and a linear transition is used in between.

2.3 The Volumetric Bull’s Eye Plot

The limitations of the bull’s eye plot mentioned in Section 2.2.1 are overcome by the volumetric bull’s eye plot; a three-dimensional extension of the bull’s eye plot that is continuous, provides transmurality information and is annotated with anatomical information. An example of a volumetric bull’s eye plot is shown in Figure 2.3b.

The volumetric bull’s eye plot is an unfolding of the myocardium along the long axis. This unfolding is illustrated in Figure 2.4 and discussed in detail in Section 2.3.1. The bottom of the cylinder corresponds to the epicardium and the top corresponds to the endocardium. The volumetric nature of the left ventricular wall is preserved. Similar to the bull’s eye plot, the contours are reformatted to concentric rings to mask any shape

(35)

The Volumetric Bull’s Eye Plot 21

0%

17 33 50 67 83 100%

(a) (b)

Figure 2.3: A comparison between (a) a bull’s eye plot and (b) a volumetric bull’s eye plot. The rings in (a) represent the reformatted contours. These are not present in (b), as the volumetric bull’s eye plot is a continuous volume. The empty rings correspond to late enhancement slices where the myocardium was not present or could not be identified. In both figures blue corresponds to healthy tissue and orange corresponds to scar.

variations. There are however two important differences, that, combined with the additional anatomical context, solve the three aforementioned disadvantages of the bull’s eye plot.

The first difference is that, as shown in Figure 2.4, the distance between the epicardium and endocardium is mapped to the thickness of the cylinder. This means that the volumetric nature of the left ventricular wall is preserved. The transmurality information, which is lost in the bull’s eye plot, is thus preserved in the volumetric bull’s eye plot.

The second difference is that there are no discontinuities between the contours since the volumetric bull’s eye plot is a continuous volume.

Therefore there is also no need for quantization.

The addition of anatomical information solves the last disadvantage.

As shown in Figure 2.3b, the coronary arteries are mapped on top of the epicardial side of the volumetric bull’s eye plot. This allows scar to be related to coronary arteries. It also allows the orientation of the volumetric bull’s eye plot with respect to the heart to be perceived. Since a late enhancement scan exhibits too little anatomical detail, the centerlines of the three primary coronary arteries are extracted from a whole heart scan of the same patient using a currently experimental prototype [50]. As additional orientation aid, two dots are drawn on the side of the volumetric bull’s eye plot that indicate the locations where the right ventricle joins the left ventricle.

(36)

short axis

h

short axis

r

(a)

(b)

Figure 2.4: A volumetric bull’s eye plot is constructed by unfolding and reformatting the myocardium to a cylinder, where the volumetric nature of the ventricle wall is preserved. Figure 2.4b shows a volume rendering of segmented late enhancement data in both original and unfolded form.

(37)

The Volumetric Bull’s Eye Plot 23 2.3.1 Unfolding the Myocardium

A parameterization of the left ventricle based on cylindrical coordinates is used for the unfolding of the myocardium. As illustrated in Figure 2.4, each point in either the myocardium or the volumetric bull’s eye plot can be represented by a point in parameterized form~x_parm = (ϕ,h,r)_{. In} this parameterization ϕ represents the angle with the short axis, which is mapped to the positive y-axis in the volumetric bull’s eye plot, pointing upward in Figure 2.3b. The variable h represents the distance to the long axis in the myocardium, which is mapped to the distance to the center of the cylinder in the volumetric bull’s eye plot. Finally, the variable r represents the distance to the long axis in the myocardium. In the volumetric bull’s eye plot it represents the displacement along the cylinder axis. Given values for ϕ and h, the difference between r of the endo- and epicardium corresponds to the wall thickness, which varies throughout the left ventricle. This wall thickness is normalized in Figure 2.3b, to obtain a more uniform appearance. More on this matter is discussed in Section 2.3.4.

The contours segmenting the myocardium are sampled non-uniformly to compensate for distortions. Let f(p)denote the angle between the short axis and a line through both p and the point on the long axis closest to p. The contours are defined as piece-wise cubic Bézier splines C(α) with α∈ [0, 2π]. These splines are sampled such that f(C(α)) =αfor any α∈ {^i2π_N : 0≤i< N}and a predefined integerN. For anyα∈ (^i2π_N ,⁽ⁱ⁺_N¹⁾^2π) with 0 ≤ i < N, C(α) is sampled to have constant chord length. This approach minimizes distortions due to the not perfectly circular shape of the contours while maintaining an accurate mapping of angles to the short axis.

Given a point in Cartesian space in the volumetric bull’s eye plot, its parameterized form can be computed using simple geometry. This parameterized form can then be directly interpreted as being in the parameterized space of the myocardium. Finally, the point in Cartesian space of the late enhancement scan can be found by using h to determine the slice the point is on, r to determine whether the point is on the endocardium or the epicardium, and ϕ as an offset in the contour corresponding to that position. Since all parameters are continuous, an interpolation scheme is required to combine multiple neighboring contours. Here Catmull-Rom interpolation is applied for ϕand h, while linear interpolation is applied forr. To minimize sampling errors, this unfolding of the myocardium is performed on-the-fly.

The on-the-fly unfolding of the myocardium is realized by computing two two-dimensional transformation maps; one for the endocardium and one for the epicardium. Each transformation map contains at each location the coordinates of the corresponding point in unfolded space. Using

(38)

transformation maps does not require the volume data to be resampled into an unfolded representation, but still provides high performance. Apart from being more memory efficient, using transformation maps does not introduce resampling artifacts.

The two transformation maps define functions f_endo : R² → _R³ _and f_epi : R² → _R³ from parameterized space to Cartesian space of the late enhancement scan. These two functions can be combined using the assumption of linearity to define a function f : R³ → _R³ given a point

~x_parm = (ϕ,h,r)as follows.

f(~x_parm) = (1−r)·f_endo(ϕ,h) +r·f_epi(ϕ,h)

Each transformation map is implemented as a two-dimensional texture, storing the value of f_endo(x,y)_and f_epi(x,y), respectively, at location(x,y)_. The required resolution of these textures depends on the resolution of the late enhancement scan. Note that since each side of the cylinder of the volumetric bull’s eye plot is circular and textures are square, the respective functions are not defined for each point on of the texture.

Linearity is assumed between the endo- and epicardial layers, as the circles, to which the contours of these layers are mapped in the volumetric bull’s eye plot, only differ in displacement along the primary axis of the cylinder. Due to this assumption of linearity, only two two-dimensional transformation maps are required instead of a full three-dimensional map.

While it may seem better to unfold along the normal direction of the center surface (in between the endo- and epicardial layers), the approach presented here preserves the relation between the distance to the center in the volumetric bull’s eye plot and the distance to the apex along the long axis in the three-dimensional case. In other words, all points on a concentric cylinder in the volumetric bull’s eye plot lie in a plane orthogonal to the long axis. This is expressed by the gray circles on the ventricle and the volumetric bull’s eye plot in Figure 2.4. This is an important and intuitive relation, which is also present in the bull’s eye plot.

2.3.2 Overall Distribution of Scar

When the cylinder that is the result of the volumetric bull’s eye plot unfolding is viewed from the bottom, it gives a similar global overview as a bull’s eye plot. This can be seen by comparing Figures 2.3a and 2.3b. The volumetric nature of this cylinder requires an algorithm that projects the three-dimensional volume onto a two-dimensional images.

In this work both direct volume rendering [46] and maximum intensity projection [87] are applied. In Figure 2.3b a direct volume rendering strategy is applied. A linear gradient of two visually distinct colors is used to color-code the degree of scar according to the scar classification. In

(39)

The Volumetric Bull’s Eye Plot 25

(a) (b)

Figure 2.5: Two volumetric bull’s eye plots that show only a slab of the cylinder; (a) shows only viability information near the epicardial layer and (b) only near the endocardial layer.

Figure 2.3 blue and yellow represent healthy tissue and scar, respectively.

More colors provide little advantage and may cause visual clutter, an overly complex appearance that hampers effective interpretation, especially when additional annotations are added.

In Figure 2.3b, all the data from the myocardial segmentation are used, similar to the bull’s eye plot. The volumetric aspect of the volumetric bull’s eye plot can be used to control the amount of data visualized by selecting a slab of the cylinder. This concept is demonstrated in Figure 2.5.

By varying the slab of the cylinder that represents the selection, the viability information either near the epicardial surface (see Figure 2.5a), the endocardial surface (see Figure 2.5b), or in the center can be visualized separately. This functionality is useful for assessing the severity and the transmurality of local scar.

2.3.3 Transmurality of Scar

The main advantage of the volumetric bull’s eye plot is its volumetric nature preserves information on the transmurality of scar, i.e. the extent of scar tissue into the left ventricular wall. This is demonstrated in Figure 2.6, where the volumetric bull’s eye plot is shown from the side with a wedge cut-out. Along the edge of the cut-out, the transmurality can be observed.

However, only a small part of Figure 2.6 actually shows the transmurality of the region of interest. The cylindrical shape of the volumetric bull’s eye

(40)

Figure 2.6: A volumetric bull’s eye plot with a wedge cut-out, exposing the transmurality of scar.

plot also makes it difficult to assess the transmurality of a larger region of interest.

Therefore a second unfolding of the myocardium is introduced specifically intended for exploring the transmurality of a region of interest. For this unfolding, called the rectangular unfolding, the myocardium is first cut open along the long axis and is then unfolded and reformatted to a cuboid.

This is illustrated in Figure 2.7. Proportions inside the myocardium are better preserved in the rectangular unfolding than in the volumetric bull’s eye plot since less stretching is involved during the reformatting. The rectangular unfolding is implemented similarly to the volumetric bull’s eye plot unfolding. The main difference is that the two transformation maps have a rectangular shape instead of circular one. When the rectangularly unfolded myocardium is viewed from the side, the transmurality can be perceived easily.

The motivation behind constructing this rectangular unfolding is as follows. On the volumetric bull’s eye plot the user can select a pie- shaped area of interest (see Figure 2.9a), for which a visualization of the transmurality is desired. The key observation is that this pie-shaped area in the volumetric bull’s eye plot corresponds to a rectangular slab in the rectangular unfolding. By visualizing only this slab and viewing it from the side, the transmurality of the region of interest can be observed (see Figure 2.9b). When the pie-shaped section would be viewed from the side, distortions would appear due to variations in thickness along the viewing direction and due to its cylindrical shape. These distortions are no longer present in the side projection of the rectangular slab.

Similar to the volumetric bull’s eye plot, this approach yields a three- dimensional object and thus requires a projection algorithm. Figure 2.9b demonstrates both a direct volume rendering and a maximum intensity projection scheme. Since the transmurality varies throughout the ventricle, it varies along the viewing direction of the slab. To obtain an accurate visualization of the transmurality, the slab should remain thin. Thus the

Comprehensive Visualization of Cardiac MRI Data

DISSERTATION

Comprehensive Visualization of Cardiac MRI Data

Comprehensive Visualization of Cardiac MRI Data

Maurice Termeer

Abstract

Kurzfassung

Contents

Preface

T

Introduction 1

1.1 The Human Heart

T

1.2 Coronary Artery Disease

1.3 Magnetic Resonance Imaging

1.4 Visualization of Cardiac MRI Data

1.5 Scope of this Thesis

The Volumetric Bull’s Eye Plot 2

2.1 Introduction

T

2.2 Related Work

2.3 The Volumetric Bull’s Eye Plot